gdiamos

7 ms·

61.
by gdiamos 8mo ago
I know the frontier “labs” are holding back publications. I don’t think it will last among researchers who think beyond production LLMs
62.
by gdiamos 8mo ago
I remember it being less profitable than graphics for a long time. It did make money that would be interesting to a startup, but not to a public company.
63.
by gdiamos 8mo ago
I remember it differently. CUDA was built with the intention of finding/enabling something like deep learning. I thought it was unrealistic too and took it on faith in people more experienced than me, until I saw deep learning work. So
64.
by gdiamos 8mo ago
Most people don't appreciate how many dead end applications NVIDIA explored before finding deep learning. It took a very long time, and it wasn't luck.
65.
by gdiamos 8mo ago
I'm not sure why the article dismisses cost. Let's say X=10% of the GPU area (~75mm^2) is dedicated to FP32 SIMD units. Assume FP64 units are ~2-4x bigger. That would be 150-300mm^2, a huge amount of area that would increase the p
66.
by gdiamos 8mo ago
I feel like a lot of software engineering problems come out of people who refuse to talk to each other than through comments in VCS. It makes sense if you are collaborating over IRC, but I feel the need to face palm when people sitting next
67.
by gdiamos 9mo ago
It sounds like lack of security is the biggest feature and risk of this clawd thing. I also tried using Siri to tell me the weather forcast while I was driving to the park. It asked me to auth into my phone. Then it asked me to approve loca
68.
by gdiamos 9mo ago
Here’s another perspective. Developers aren’t budget approvers in engineering organizations by choice. If you are a budget approver then your inbox and calendar are full of sales teams. I find that experience is too distracting to concentra
69.
by gdiamos 9mo ago
It was an interesting read Sara, thanks for sharing it. I especially agree with your point that scaling laws really killed open research. That's a shame and I personally think we could benefit from more research. I originally didn&#x27
70.
by gdiamos 11mo ago
The iphone is so useful they can probably ride it for a couple more decades at least. I would still buy it. From a technology or engineering perspective, I have no idea how to work with Apple.
71.
by gdiamos 11mo ago
A Dyson sphere brain ?
72.
by gdiamos 11mo ago
I personally don’t think the scaling hypothesis is wrong, but it is running up against real limits What high quality data sources are not already tapped? Where does the next 1000x flops come from?
73.
by gdiamos 11mo ago
Diffusion is favored by current GPUs . Over time we seem to have a tendency to build models that are well matched to our machines
74.
by gdiamos 11mo ago
How do they break ties?
75.
by gdiamos 11mo ago
Did I miss something? https://github.com/NVlabs/Fast-dLLM/blob/main/llada/chat.py That’s inference code, but where is the high perf web server?
76.
by gdiamos 11mo ago
Thanks I’ll check it out!
77.
by gdiamos 11mo ago
One year later and there is still no inference engine for diffusion LLMs Students looking for a project to break into AI - please!
78.
by gdiamos 11mo ago
It was risky. From my perspective it was more exciting to the programming systems and compiler community than to the computer architecture community.
79.
by gdiamos 11mo ago
Transmeta made a technology bet that dynamic compilation could beat OOO super scalar CPUs in SPEC. It was wrong, but it was controversial among experts at the time. I’m glad that they tried it even though it turned out to be wrong. Many of
80.
by gdiamos 11mo ago
The role of basic research is to get off the beaten path. LLMs aren’t basic research when they have 1 billion users
81.
by gdiamos 11mo ago
What is going on at meta? Soumith probably knew about Lecun. I’m taking a second look at my PyTorch stack.
82.
by gdiamos 11mo ago
This is the end of an era. Amazing work soumith.
83.
by gdiamos 11mo ago
ScalarLM uses tokenformer adaptors by default, which have learnable key/values https://www.scalarlm.com/blog/tokenformer-a-scalable-transfo...
84.
by gdiamos 1y ago
> An expert human compiler... yes, that would be SufficientlySmart. > And slow. Build faster GPUs. > And expensive. The more you buy the more you save. > And prone to error. Find ways to keep scaling laws bringing down the error
85.
by gdiamos 1y ago
pip culture
86.
by gdiamos 1y ago
I find it helpful to read a saxpy and GEMM kernel for a new accelerator like this - do they have an example?
87.
by gdiamos 1y ago
Reminds me of the movie memento
88.
by gdiamos 1y ago
This is exactly the intuition that leads to excitement about fine tuning. However, I personally think that this intuition applies to products and interfaces, not to AI. Intelligence and learning is general. Intelligence without generalizati
89.
by gdiamos 1y ago
The problem is that it doesn’t always work and when it does fail it fails silently. Debugging requires knowing some small detail about your data distribution or how you did gradient clipping which take time and painstakingly detailed experi
90.
by gdiamos 1y ago
I agree this is the main case where it makes sense. But a recent trend that cut into the cost savings is that foundation model companies have started releasing small models. So you can build a use case with qwen 235B, then shrink down to 30

More ›